March test algorithm for unlinked static reduced three-cell coupling faults in random-access memories P. Cas ¸caval a, * , D. Cas ¸caval b a “Gheorghe Asachi” Technical University of Ias ¸i, Department of Computer Science and Engineering, Bd. D. Mangeron, nr. 27, Ias ¸i, 700050, Romania b “Gheorghe Asachi” Technical University of Ias ¸i, Department of Industrial Engineering, Bd. D. Mangeron, nr. 27, Ias ¸i, 700050, Romania ABSTRACT A memory fault model regarding the unlinked static three-cell coupling faults in n  1 random-access memories is discussed. This model is an extension of the well- known model of unlinked static two-cell coupling faults. Because this model of three-cell coupling is limited to the physically neighbouring memory cells, it can also be considered a neighbourhood pattern-sensitive model. An efficient march test algorithm able to cover this reduced model of three-cell coupling is presented in this letter. 1. Introduction The fault model of unlinked static three-cell coupling faults in n  1 random-access memories as presented in Ref. [1] is discussed. Based on the model of all static simple two-cell coupling faults pre- sented in Ref. [2], a fault primitive (FP) based-model of three-cell coupling is defined in Ref. [1]. A set of 72 FPs completely covers this model of three-cell coupling faults. It is said that two or more FPs are not linked when they do not in- fluence each other. According to the fault classification presented in Refs. [2,3], the class of ‘static faults’ refers to those faults which are sensitized by performing at most one operation in the memory, whereas the class of ‘dynamic faults’ refers to those faults which can be sensitized by per- forming more than one operation sequentially. As in Ref. [1], we only address the class of static coupling faults. The first test algorithm dedicated to a model of three-cell coupling is proposed by Nair, Thatte and Abraham [4]. Other two more efficient test algorithms are given by Cockburn (S3CTEST and S3CTEST2) [5]. Because the authors assume that the coupling cells can be anywhere in the memory, all these test algorithms are quite long. For example, the test algorithm S3CTEST needs 5n log 2 n þ 22:5n operations. To reduce the length of the tests, Cas ¸caval and Bennett [6] have limited to the more realistic case where only the physically neighbouring memory cells may form a three-cell coupling. For this model, they proposed a march memory test algorithm (MT) with a length of 38n. An improved algo- rithm dedicated to this model (MT-R3CF) with a length of 30n is given by Cas ¸caval, Bennett and Hut ¸anu [7]. Both test algorithms, MT and MT-R3CF, assume that a memory fault can be sensitized only by a transition write operation into a cell. In Ref. [1], the model of three-cell coupling is extended by considering other classes of faults, such as the faults sensitized by a non-transition write or a read operation, namely: disturb coupling, read destructive coupling, deceptive read destructive coupling or incorrect read, as defined in Ref. [2] and in other works. This is the model we consider in this work. As in Refs. [1,6,7], the authors restricted their study to the case where only the physically neighbouring memory cells may be affected by a three-cell coupling fault. As in Ref. [1], six coupling patterns of three physically neighbouring cells (denoted by CP 1 , CP 2 , …, CP 6 ) are considered in this work, as presented in Fig. 1. This model is known as ‘reduced three-cell coupling’. For a better comparison, all the preliminary considerations presented in Ref. [1] are also accepted in this work. Note that the model we discuss can also be viewed as a neighbourhood pattern-sensitive fault model (NPSF). Nevertheless, in this model, any cell in the group may be a victim cell of the other two aggressor cells, not just the central cell, as it is usually considered in the NPSF model. As in any work dedicated to NPSFs, we assume that the scramble map is completely known, so we can run the test using this physical address information. Naturally, in the memory under testing, one or more groups of coupled cells may exist. As in Ref. [1–7], we assume that the groups of coupled cells are disjoint. For this model, the test algorithm March SR3C with a length of 66n operations is proposed in Ref. [1]. In this letter, we present a more effi- cient march test (MT-SR3C) with only 54n operations able to cover this model of unlinked static three-cell coupling faults. Notation: The following notations are used to describe a memory test algorithm: * Corresponding author. E-mail addresses: cascaval@cs.tuiasi.ro (P. Cas ¸caval), cascaval@tex.tuiasi.ro (D. Cas ¸caval). Contents lists available at ScienceDirect Microelectronics Journal journal homepage: www.elsevier.com/locate/mejo https://doi.org/10.1016/j.mejo.2019.104619 Received 24 March 2019; Received in revised form 27 July 2019; Accepted 16 September 2019 Available online 19 September 2019 0026-2692/© 2019 Elsevier Ltd. All rights reserved. Microelectronics Journal 93 (2019) 104619